Using with AI (MCP Guide)

The Model Context Protocol (MCP) allows AI coding assistants to talk directly to tools like CodeGraphContext.

1. Run the MCP Setup Wizard

We provide an interactive tool to configure your editors automatically.

cgc mcp setup

What happens here:

1. The tool looks for configuration files (e.g., ~/Library/Application Support/Cursor/User/globalStorage/mcp.json).
2. It injects the CodeGraphContext server details.
3. It ensures the server knows how to find your database.

2. Supported Clients

Client	Setup Method	Notes
Cursor	Automatic	Requires "MCP" feature enabled in settings.
Claude Desktop	Automatic	Works with the Claude 3.5 Sonnet model.
VS Code	Semi-Automatic	Requires the "Continue" extension or similar MCP client.

3. How to Use It (Once Connected)

Open your AI Chat and talk naturally. The AI now has a "tool" it can call.

Example Prompts:

• "Please index the current directory." -> AI calls add_code_to_graph
• "Who calls the process_payment function?" -> AI calls analyze_callers
• "Find all dead code in utils.py." -> AI calls find_dead_code

4. Troubleshooting

• "Component not found": This usually means the MCP server didn't start. Check the logs in your AI editor.
• "Database error": Ensure your Neo4j container is running (docker ps) or that your Python environment is active.